【摘要】 凭借测量精度高、速度快、可测范围大、非接触等优点，多目视觉测量系统在航空航天、汽车等领域的动态目标高精度空间定位中应用广泛，但现有基于光束法平差的多目视觉测量系统的三维重建精度和速度仍无法满足动态目标空间位置的高精度快速测量需求。因此，提出了一种基于分步光束法平差的多目视觉测量系统重建算法，分步优化了不同数量级的系统参数，并通过FPGA平台实现了重建算法的快速运行。实验结果表明，本文提出的分步光束法平差重建算法的平均空间重投影误差小于68μm，优于直接线性变换重建算法与传统光束法平差重建算法的精度。该FPGA硬件架构方案可同时对4路高清图像数据（2 048×2 048×8 bit）实现约40 frame/s的处理速度，满足动态目标空间位置的实时测量需求。更多还原

【Abstract】 Due to its high measurement accuracy, rapid acquisition speed, large measurable range, and non-contact characteristics, the multi-camera vision measurement system has been widely applied to highprecision spatial positioning of dynamic targets in aerospace, automotive, and related fields. However, existing multi-camera vision measurement systems based on bundle adjustment cannot simultaneously satisfy the stringent requirements of high accuracy and high speed in three-dimensional reconstruction of dynamic target positions. In this study, a reconstruction algorithm for multi-camera vision measurement systems based on stepwise bundle adjustment is proposed. System parameters with different orders of magnitude are optimized sequentially, and the rapid execution of the reconstruction algorithm is implemented on an field-programmable gate array(FPGA) platform. Experimental results demonstrate that the proposed stepwise bundle adjustment reconstruction algorithm achieves an average spatial reprojection error of less than 68 μm, outperforming both the direct linear transformation reconstruction algorithm and the conventional bundle adjustment reconstruction algorithm. The designed FPGA hardware architecture attains a processing speed of approximately 40 frame/s for four channels of high-definition image data(2 048×2 048×8 bit) simultaneously, thereby meeting the real-time, high-precision measurement requirements for the spatial positioning of dynamic targets.更多还原